[Refactor] Remove dead code for quantization (#52221)

Signed-off-by: yewentao256 <zhyanwentao@126.com>
This commit is contained in:
Wentao Ye
2026-08-16 13:10:16 -04:00
committed by GitHub
parent 6914d60b1e
commit 7d7b6f26f4
14 changed files with 1 additions and 395 deletions
+1 -23
View File
@@ -89,28 +89,6 @@ class FP8Params(Params):
)
@dataclass
class Int8Params(Params):
"""Int8 layer parameters with typed fields"""
input_zero_point: torch.Tensor | None
azp_adj: torch.Tensor | None
INPUT_ZERO_POINT: ClassVar[str] = "input_zero_point"
AZP_ADJ: ClassVar[str] = "azp_adj"
@classmethod
def from_layer(cls, layer: torch.nn.Module) -> "Int8Params":
"""Extract parameters from layer"""
return cls(
weight=getattr(layer, cls.WEIGHT),
weight_scale=getattr(layer, cls.WEIGHT_SCALE),
input_scale=getattr(layer, cls.INPUT_SCALE, None),
input_zero_point=getattr(layer, cls.INPUT_ZERO_POINT, None),
azp_adj=getattr(layer, cls.AZP_ADJ, None),
)
_ParamsT = TypeVar("_ParamsT", bound=Params)
_ConfigT = TypeVar("_ConfigT", bound=MMLinearLayerConfig)
@@ -130,7 +108,7 @@ class MMLinearKernel(ABC, Generic[_ConfigT, _ParamsT]):
Typical Usage:
1. Define a config dataclass inheriting from MMLinearLayerConfig
2. Define a params dataclass inheriting from Params (or FP8Params/Int8Params)
2. Define a params dataclass inheriting from Params
3. Subclass MMLinearKernel with your config and params types
4. Implement all abstract methods
5. Register the kernel with the quantization method